termNN

CNN- and LSTM-based models for detection of intrinsic transcription terminators and tRNAs in bacteria.

package requirements

tensorflow v2.1
h5py v2.10.0
numpy v1.19.5

train and validate models for terminators + tRNAs

models_training.py: Pretrain and train CNN and LSTM models.
models_validating.py: Validate all trained models, write output to results/validation, plot results.

analyse the impact of structure of terminators + tRNAs, and the impact of strcuture of terminators

mutate_basepairs.py: Mutate an increasing number of base pairs in tRNAs and terminators, feed them into all trained models, write output to results.
mutate_sections.py: Mutate half of the nucleotides of each section of terminators, feed them into all trained models, write output to results.
mutate_processing.py: Summarize results of the mutation experiments, write output to results/mutation, plot results.

scan genome for terminators

scan_genome.py: Use with one or more options of ['encode_genome', 'run_scan', 'find_kernels', 'calc_PPV']:

• option 'encode_genome': Splits the input genome into overlapping subsequences of 75 base pairs and encodes them for the scan. The resulting encoded sequences are written to directory genomescan/encoded_genome.
• option 'run_scan': Runs the actual scan. Reads encoded sequences from genomescan/encoded_genome and writes all model output of all models to genomescan/results_single_nts.
• option 'find_kernels': searches for stretches of nucleotides with a model output greater 0.5 and combines them to one kernel. Reads model output from genomescan/results_single_nts, writes results to genomescan/results_kernels.
• option 'calc_PPV': Calculates PPV for genome scan.